Big Data Analysis and Artificial Intelligence for Medical Sciences

Bruno Carpentieri is Associate Professor in the Faculty of Engineering at the Free University of Bozen-Bolzano, Bozen-Bolzano, Italy.Paola Lecca is Assistant Professor in the Faculty of Engineering at the Free University of Bozen-Bolzano, Bozen-Bolzano, Italy.

• List of Contributors xiiiPreface xix1 Introduction 1Bruno Carpentieri and Paola Lecca1.1 Disease Diagnoses 41.2 Drug Development 61.3 Personalized Medicine 61.4 Gene Editing 7Author Biographies 9References 92 Fuzzy Logic for Knowledge-Driven and Data-Driven Modeling in Biomedical Sciences 17Paolo Cazzaniga, Simone Spolaor, Caro Fuchs, Marco S. Nobile and Daniela Besozzi2.1 Introduction 172.2 Fuzzy Logic 182.2.1 Fuzzy Sets 192.2.2 Linguistic Variables 192.2.3 Fuzzy Rules 202.2.4 Fuzzy Inference Systems 212.2.5 Simpful 222.3 Knowledge-Driven Modeling 222.3.1 Dynamic Fuzzy Modeling 232.3.2 Application 1: Maximizing Cancer Cells Death with Minimal Drug Combinations 252.3.3 FuzzX: A Hybrid Mechanistic-Fuzzy Modeling and Simulation Engine 272.3.4 Application 2: Analyzing Oscillatory Regimes in Signal Transduction Pathways 292.4 Data-Driven Modeling 302.4.1 pyFUME: Automatic Generation of Fuzzy Inference Systems 312.4.2 Application 3: Assessing Tremor Severity in Neurological Disorders 332.5 Discussion 35Author Biographies 36References 373 Application of Machine Learning Algorithms to Diagnosis and Prognosis of Chronic Wounds 43Mai Dabas and Amit Gefen3.1 Background 433.1.1 Chronic Wounds 433.1.2 Implementation of AI Methodologies in Wound Care and Management 433.2 Clinical Visual Assessment of Wounds Supported by Artificial Intelligence 443.2.1 Predicting the Formation and Progress of Wounds Based on Electronic Health Records 463.2.2 Predicting the Formation and Evolution of Wounds Based on a Dynamic Evaluation of Wound Characteristics and Relevant Physiological Measures 483.2.3 Feasible Implementation of AI Solutions For Wound Care Delivery and Management 493.2.4 Types of Data Modalities for Diagnosis, Detection, and Prediction of Chronic Wounds 503.3 Smartphone and Tablet Use in Wound Diagnosis and Management 513.4 Conclusions 53Acronyms 54Author Biographies 55References 554 Deep Learning Techniques for Gene Identification in Cancer Prevention 59Eleonora Lusito4.1 The Next-Generation Era of Cancer Investigation 594.1.1 Cancer at Its First Definitions 594.1.2 Attempts to Sequence Nucleic Acids Over the Years 604.1.3 From the First to the Third-Generation Sequencing 614.1.4 Applications of NGS in Clinical Oncology 624.2 Deep Learning Approaches for Genomic Variants Identification in Cancer 634.2.1 Cancer Causing Factors 634.2.2 The Contribution of Germline Alterations to Cancer 644.2.3 Somatic Mutations and Cancer 644.2.4 Calling Variants from Sequence Data 654.2.5 Computational Approaches for Variant Discovery 654.2.6 Convolutional Neural Networks (CNNs): Basic Principles 664.2.7 Application of CNNs to Variant Calling 674.2.8 A Typical CNN Architecture for Variant Calling 684.2.9 The Activation Function 694.2.10 Dropout and L1–L2 Regularization 714.2.11 Advantages of Deep Learning Over the Existing Techniques 724.2.12 Residual Neural Networks (ResNet)-Inspired CNN in Genomic Variants Detection 734.3 Deep Learning in Cancer Transcriptomics 744.3.1 Gene Expression and Cancer 744.3.2 Analytical Approaches to Deal with Gene Expression Data 764.3.3 Stacked Denoising Autoencoders (SDAEs) for Dimensionality Reduction 764.3.4 The Variational Autoencoder (VAE) 794.3.5 VAEs to Integrate Gene Expression and Methylation Data 814.3.5.1 DNA Methylation: the Epigenetic Regulation of Gene Expression 814.3.5.2 Preprocessing Input Data of Different Sources 824.3.5.3 A VAE Architecture for Multimodal Data 824.4 Conclusions 84Acronyms 86Author Biographies 87References 875 Deep Learning for Network Biology 97Eleonora Lusito5.1 Types of Interactions Between Genes and Their Products 975.2 Deep Learning Methods with Graph-input Data 995.2.1 Graph Embedding 995.2.1.1 Random Walk-Based Graph Embedding 1005.2.1.2 Proximity-Based Graph Embedding 1015.2.2 Graph Convolutional Networks (GCNs) 1025.3 Applications of GNNs to Infer Biological and Pharmacological Interactions 1045.3.1 Proteomics 1045.3.2 Drug Development and Repurposing 1045.3.3 Drug–Drug Interaction Prediction 1055.3.4 Disease Classification and Outcome Prediction 106Author Biography 107References 1076 Deep Learning-Based Reduced Order Models for Cardiac Electrophysiology 115Stefania Fresca, Luca Dedè and Andrea Manzoni6.1 Overview of Cardiac Physiology 1156.1.1 Atrial Tachycardia and Atrial Fibrillation 1176.1.2 Mathematical Models for Cardiac Electrophysiology 1186.2 Reduced Order Modeling 1216.2.1 Problem Formulation 1236.2.2 Nonlinear Dimensionality Reduction 1236.3 Decreasing Complexity in Cardiac Electrophysiology 1246.3.1 POD-Enhanced Deep Learning-Based ROMs 1256.3.1.1 POD-DL-ROM Architecture and Algorithms 1286.4 Numerical Results 1306.4.1 Test 1: Two-Dimensional Slab with Figure of Eight Reentry 1316.4.2 Test 2: Three-Dimensional Left Ventricle Geometry 1336.4.3 Test 3: Left Atrium Surface by Varying the Stimuli Location 1356.4.4 Test 4: Reentry Breakup 1376.5 Conclusions 139Author Biographies 140References 1407 The Potential of Microbiome Big Data in Precision Medicine: Predicting Outcomes Through Machine Learning 149Silvia Turroni and Simone Rampelli7.1 The Gut Microbiome: A Major Player in Human Physiology and Pathophysiology 1497.2 Machine Learning Applied to Microbiome Research 1517.2.1 Case Study 1: Obesity 1517.2.2 Case Study 2: Cancer 1537.2.3 Case Study 3: Personalized Nutrition 1547.2.4 Case Study 4: Exploiting the Meta-Community Theory for New Machine Learning Approaches 1557.3 Conclusions and Perspectives 155Author Biographies 156References 1568 Predictive Patient Stratification Using Artificial Intelligence and Machine Learning 161Thanh-Phuong Nguyen, Thanh T. Giang, Quang T. Pham and Dang H. Tran8.1 Overview of Artificial Intelligence for Patient Stratification 1618.2 A RPCA and MKL Combination Model for Patient Stratification 1648.2.1 Robust Principal Component Analysis 1648.2.2 Dimensionality Reduction and Features Extraction Based on RPCA 1668.2.3 Predictive Model Construction Based on Multiple Kernel Learning 1688.2.4 Materials 1698.2.4.1 Cancer Patient Datasets 1698.2.4.2 Alzheimer Disease Patient Datasets 1708.2.5 Experiment Design 1718.2.5.1 Experiment of Stratifying Cancer Patients 1718.2.5.2 Experiment of Stratifying Alzheimer Disease Patients 1718.2.6 Results and Discussions 1718.2.6.1 Application of Stratifying Cancer Patients 1728.2.7 Application of Stratifying Alzheimer Disease Patients 1748.3 Conclusion 175Author Biographies 175References 1769 Hybrid Data-Driven and Numerical Modeling of Articular Cartilage 181Seyed Shayan Sajjadinia, Bruno Carpentieri and Gerhard A. Holzapfel9.1 Introduction 1819.2 Knee and Cartilage 1829.2.1 Main Joint Substructures 1829.2.2 Load-Bearing Cartilage Phases 1839.3 Physics-Based Modeling 1859.3.1 Numerical Modeling 1859.3.2 Constitutive Modeling 1889.4 AI-Enhanced Modeling 1919.4.1 Deep Learning 1919.4.2 Surrogate Modeling 1929.5 Discussion and Conclusion 194Author Biographies 194References 19510 A Hybrid of Differential Evolution and Minimization of Metabolic Adjustment for Succinic and Ethanol Production 205Zhang N. Hor, Mohd S. Mohamad, Yee W. Choon, Muhammad A. Remli and Hairudin A. Majid10.1 Introduction 20510.2 Method 20610.2.1 Differential Evolution (DE) 20610.2.2 Mutation 20610.2.3 Crossover 20710.2.4 Selection 20810.2.5 Minimization of Metabolic Adjustment 20810.2.6 A Hybrid of Differential Evolution and Minimization of Metabolic Adjustment 20910.3 Experiments and Discussion 20910.3.1 Dataset 20910.3.2 Parameter Setting 20910.3.3 Experimental Results 21010.3.4 Comparative Analysis 21410.4 Conclusion 214Acknowledgment 215Author Bibliographies 215References 21611 Analysis Pipelines and a Platform Solution for Next-Generation Sequencing Data 219Víctor Duarte, Alesandro Gómez and Juan M. Corchado11.1 Introduction 21911.2 NGS Data Analysis Pipeline and State of the Art Tools 22011.2.1 Quality Assessment 22011.2.2 Alignment 22111.2.3 Post-alignment and pre-variant Calling Processing 22211.2.4 Variant Calling 22311.2.5 Variant Annotation 22811.3 Nanopore Sequencing Data Analysis 22911.3.1 Base-Calling 23011.3.2 Quality Control and Preprocessing 23011.3.3 Error Correction 23111.3.4 Alignment 23111.3.5 Variant Calling 23111.4 Machine Learning Approaches in Variant Calling 23211.5 Next-Generation Sequencing Data Analysis Frameworks 23311.6 DeepNGS 23511.6.1 Pipeline 23511.6.2 DeepNGS Main Features 23611.6.2.1 Power and Speed 23611.6.2.2 Optimized Workflow 23611.6.2.3 Intuitive Design and Interactive Charts 23711.6.2.4 Extended Information 23711.6.2.5 Artificial Intelligence and Machine Learning 23711.7 Conclusions 240Author Biographies 241References 24112 Artificial Intelligence: From Drug Discovery to Clinical Pharmacology 253Paola Lecca12.1 Artificial Intelligence and the Druggable Genome 25312.2 Feature-Based Methods 25712.3 Similarity/Distance-Based Methods 25712.4 Matrix Factorization 25812.4.1 Causal K-Nearest-Neighborhood 26112.4.2 Causal Random Forests 26312.4.3 Causal Support Vector Machine 26412.5 Opportunities and Challenges 265Author Biography 266References 26613 Using AI to Steer Brain Regeneration: The Enhanced Regenerative Medicine Paradigm 273Gabriella Panuccio, Narayan P. Subramaniyam, Angel Canal-Alonso, Juan M. Corchado and Carlo Ierna13.1 The Challenge of Brain Regeneration 27313.2 The Enhanced Regenerative Medicine Paradigm 27413.3 The Case of Epilepsy 27613.4 AI to Understand Epilepsy 27913.4.1 Commonly Applied Learning Algorithms for Basic Neuroscience and Clinical Application in Epilepsy 28213.4.2 Seizure and Epilepsy Type Classification 28413.4.3 Seizure Onset Zone Localization 28413.4.4 Seizure Detection 28513.4.5 Seizure Prediction 28513.4.6 Signal Feature Extraction for Seizure Detection and Prediction 28813.4.7 Network Interactions and Evolving Dynamics in the Epileptic Brain: The Eye of AI 29013.5 Artificial Intelligence to Guide Graft-Host Dynamics in Epilepsy 29213.6 Challenges and Limitations 29413.6.1 From AI to Explainable AI 29513.7 A Philosophical Perspective on Enhanced Brain Regeneration 297Acknowledgments 299Acronyms 299Author Biographies 300References 30014 Towards Better Ways to Assess Predictive Computing in Medicine: On Reliability, Robustness, and Utility 309Federico Cabitza and Andrea Campagner14.1 Introduction 30914.2 On Ground Truth Reliability 31114.2.1 Weighted Reliability 31414.2.2 Example Application 31614.3 On Utility Metrics to Evaluate ML Performance 31814.3.1 Weighted Utility 31814.3.2 Example Application 32114.4 On the Replicability of Clinical ML Models 32214.4.1 Dataset Size 32314.4.2 Dataset Similarity 32514.4.3 Meta-Validation Procedure 32514.4.4 Example Application 32814.5 Conclusions and Future Outlook 331Author Biographies 332References 33315 Legal Aspects of AI in the Biomedical Field. The Role of Interpretable Models 339Chiara Gallese15.1 Introduction 33915.2 Data Protection 34015.3 Transparency Principle 34315.3.1 Right of Explanation 34315.3.2 Right of Information 34815.3.3 Informed Consent Requirements 34915.4 Accountability Principle 35015.5 Non-discrimination Principle and Biases 35115.6 High-Risk Systems and Human Oversight 35315.7 Additional Requirements of the AI Act Proposal 35415.8 Interpretability as a Standard 35515.9 Conclusion 358Author Biography 358References 35916 The Long Path to Usable AI 363Barbara Di Camillo, Enrico Longato, Erica Tavazzi and Martina Vettoretti16.1 Promises and Challenges of Artificial Intelligence in Healthcare 36316.2 Deployment of Usable Artificial Intelligence Models 36716.2.1 Case Study: Predicting the Cardiovascular Complications of Diabetes via a Deep Learning Approach 36816.3 Potential and Challenges of Employing Longitudinal Clinical Data in AI 37516.3.1 Case Study: Modeling the Progression of Amyotrophic Lateral Sclerosis Through a Dynamic Bayesian Network 37816.3.2 Case Study: Investigating Amyotrophic Lateral Sclerosis Progression Trajectories Leveraging Process Mining 38116.4 Enhancing the Applicability of AI Predictive Models by a Combined Model Approach: A Case Study on T2D Onset Prediction 38616.4.1 The Problem of Type 2 Diabetes Prediction 38616.4.2 Potential Applications of T2D Predictive Models 38716.4.3 Barriers to the Adoption of T2D Predictive Models 38716.4.4 Addressing Practical Issues by Combining Multiple T2D Predictive Models 38816.4.5 The Combined Model Achieves High Prediction Performance with High Coverage 39016.5 Conclusions and Future Outlook 391Author Biography 392References 393Index 399